New computations for RMSEA and CFI following FIML and TS estimation with missing data

Area of Science:

• Statistics
• Psychometrics
• Quantitative Psychology

Background:

• Full-information maximum likelihood (FIML) is common for missing data in structural equation modeling (SEM).
• Standard computation of SEM approximate fit indices (AFIs) with FIML can yield distorted results compared to complete data.
• Existing methods struggle with bias in AFIs when data are missing.

Purpose of the Study:

• To propose and evaluate the FIML-corrected (FIML-C) approach for computing AFIs with FIML estimation.
• To examine the two-stage (TS) approach for computing AFIs with missing data.
• To introduce small sample corrections for both FIML-C and TS approaches.

Main Methods:

• Developed and simulated the FIML-corrected (FIML-C) approach for AFIs.
• Evaluated the two-stage (TS) approach for AFIs.
• Implemented small sample corrections for both methods.
• Conducted two simulation studies under various missing data and model misspecification conditions.

Main Results:

• FIML-C and TS approaches, with small sample corrections, accurately estimated population-complete-data AFIs with minimal bias.
• The FIML-C approach showed limitations in specific conditions with high missing data and model misspecification.
• Standard FIML computations for AFIs frequently performed poorly.

Conclusions:

• The FIML-C and TS approaches are recommended for computing AFIs in SEM with missing data.
• These methods effectively mitigate bias introduced by missing data.
• Small sample corrections enhance the reliability of AFIs estimated by FIML-C and TS.